<p>Soil is a huge carbon (C) reservoir, but where and how much extra C can be stored is unknown. Here, using 5089 observations, we estimated that the uppermost 30&nbsp;cm of Australian soil holds 13&nbsp;Gt (10–18&nbsp;Gt) of mineral-associated organic carbon (MAOC). Using a frontier line analyses, described in Viscarra Rossel et al. (2023), we estimated the maximum amounts of MAOC that Australian soils could store in their current environments, and calculated the MAOC deficit, or C sequestration potential. We propagated the uncertainties from the frontier fitting and mapped the estimates of these values over Australia using machine learning and kriging with external drift (KED). The maps show regions where the soil is more in MAOC deficit and has greater sequestration potential. The modelling shows that the variation over the whole continent is determined mainly by climate, linked to vegetation, and soil mineralogy. We find that the MAOC deficit in Australian soil is 40&nbsp;Gt (25–60&nbsp;Gt). The deficit in the vast rangelands is 20.84&nbsp;Gt (13.97–29.70&nbsp;Gt) and the deficit in cropping soil is 1.63&nbsp;Gt (1.12–2.32&nbsp;Gt). Our findings suggest that the C sequestration potential of Australian soil is limited by climate.

The soil in terrestrial and blue carbon ecosystems (BCE; mangroves, tidal marshes, seagrasses) is a significant carbon (C) sink. National assessments of C inventories are needed to protect them and aid nature-based strategies to sequester atmospheric carbon dioxide. We harmonised measurements from Australia's terrestrial and BCE and, using consistent multi-scale spatial machine learning, unravelled the drivers of soil organic carbon (SOC) variation and digitally mapped their stocks. The modelling shows that climate and vegetation are continentally the primary drivers of SOC variation. But the underlying regional drivers are ecosystem type, terrain, clay content, mineralogy, and nutrients. The digital soil maps indicate that in the 0-30&nbsp;cm soil layer, terrestrial ecosystems hold 27.6&nbsp;Gt (19.6-39.0&nbsp;Gt), and BCE 0.35&nbsp;Gt (0.20-0.62&nbsp;Gt). Tall open eucalypt and mangrove forests have the largest mean SOC per unit area. Eucalypt woodlands and hummock grassland, which occupy vast areas, store the largest total SOC stock. These ecosystems constitute important regions for conservation, emissions avoidance, and preservation because they also provide additional co-benefits.

These datasets consist of soil maps generated to assess baselines, drivers and trends for soil health and stability within the NSW Regional Forest Agreement (RFA) regions. <br> The maps are organised into empirical soil maps, digital soil maps, and data cube maps. <br> Empirical soil maps consists of four products. Maps include topsoil pH, carbon, Emerson Aggregate Stability and Soil Profile Quality Confidence. Each map consists of 2,162 units. Maps were generated using the most representative soil profile for each unit available within the Soil and Land Information System (SALIS). The 2008 woody vegetation coverage was used as baseline. Maps reflect values when the sampling occurred with temporal changes not being accounted for. Locations with missing or of poor quality data are identified, providing a confidence rating map as part of the evaluation process.<br> Digital soil maps include map products of key soil condition indicators covering the Regional Forest Agreement regions of eastern NSW. Raster maps of key soil indicators, such as soil carbon, pH, bulk density, hillslope erosion and others, were created at 100 m resolution. For each key soil indicator, maps include baseline (approximately 2008) levels as well as trends of change resulting from different human and natural disturbances such as forest harvesting, uncontrolled stock grazing, climate change and bush fire. <br> Data cube maps include time series of soil organic carbon (SOC) between January 1990 and December 2020 for the Regional Forest Agreement regions of eastern NSW. Products provide estimates of SOC concentrations and associated trends through time. Modelling was carried out using a data cube platform incorporating machine learning space-time framework and geospatial technologies. Important covariates required to drive this spatio-temporal modelling were identified using the Recursive Feature Elimination algorithm (RFE). <br> A web mapping application on the NSW Spatial Collaboration Portal depicts these datasets. Access the webapp through the link below:<br> https://portal.spatial.nsw.gov.au/portal/home/item.html?id=af9c71935f024f4a8f64cb39f5eba007